On counting homomorphisms to directed acyclic graphs

Transcription

1 Electronic Colloquium on Computational Complexity, Report No. 121 (2005) On counting homomorphisms to directed acyclic graphs Martin Dyer Leslie Ann Goldberg Mike Paterson October 17, 2005 Abstract We give a dichotomy theorem for the problem of counting homomorphisms to directed acyclic graphs. H is a fixed directed acyclic graph. The problem is, given an input digraph G, how many homomorphisms are there from G to H. We give a graph-theoretic classification, showing that for some digraphs H, the problem is in P and for the rest of the digraphs H the problem is #P-complete. An interesting feature of the dichotomy, which is absent from related dichotomy results, is that there is a rich supply of tractable graphs H with complex structure. 1 Introduction Our result is a dichotomy theorem for the problem of counting homomorphisms to directed acyclic graphs. A homomorphism from a (directed) graph G = (V, E) to a (directed) graph H = (V, E) is a function from V to V that preserves (directed) edges. That is, the function maps every edge of G to an edge of H. Hell and Nešetřil gave a dichotomy theorem for the decision problem for undirected graphs H. In this case, H is an undirected graph (possibly with self-loops). The input, G, is an undirected simple graph. The question is Is there a homomorphism from G to H. Hell and Nešetřil [7] showed that the decision problem is in P if the fixed graph H has a loop, or is bipartite. Otherwise, it is NP-complete. Dyer and Greenhill [3] established a dichotomy theorem for the corresponding counting problem in which the question is How many homomorphisms are there from G to H. They showed that the problem is in P if every component of H is either a complete graph with all loops present or a complete bipartite graph with no loops present 1. Otherwise, it is #P-complete. Bulatov and Grohe [1] extended the counting dichotomy theorem to the case in which H is an undirected multigraph. Their result will be discussed in more detail below. In this paper, we consider the related counting problem for directed graphs. First, consider the corresponding decision problem. H is a fixed digraph, and the problem, given an input Partially supported by the EPSRC grant Discontinuous Behaviour in the Complexity of Randomized Algorithms. Some of the work was done while the authors were visiting the Mathematical Sciences Research Institute in Berkeley. School of Computing, University of Leeds, Leeds LS2 9JT, United Kingdom. Department of Computer Science, University of Warwick, Coventry, CV4 7AL, United Kingdom. 1 The graph with a singleton isolated vertex is taken to be a (degenerate) complete bipartite graph with no loops. 1 ISSN

2 digraph G, is Is there a homomorphism from G to H? It is conjectured [8, Conjecture 5.12] that there is a dichotomy theorem for this problem, in the sense that, for every H, the problem is either polynomial-time solvable or NP-complete. Currently, there is no graphtheoretic conjecture stating what the two classes of digraphs will look like. Obtaining such a dichotomy may be difficult. Indeed (see [Theorem 5.14][8]), Feder and Vardi have shown [5, 6] that the resolution of the dichotomy conjecture for digraphs would also resolve their longstanding dichotomy conjecture for constraint satisfaction problems. There are some known dichotomy classifications for restricted classes of digraphs, however, the problem is open [8] even when H is restricted to be an oriented tree. In this paper, we give a dichotomy theorem for the corresponding counting problem, for the class of digraphs in which H can be any acyclic directed graph. An interesting feature of this problem, which is different from the previous dichotomy theorems that we have mentioned, is that there is a rich supply of tractable graphs H with complex structure. The formal statement of our dichotomy is given below. Here is an informal description. First, the problem is #P-complete unless H is a layered digraph, meaning that the vertices of H can be arranged in levels, with edges going from one level to the next. We show (see Theorem 6.1 for a precise statement) that the problem is in P for a layered digraph H if the following condition is true (otherwise it is #P-complete). The condition is that, for every pair of vertices x and x on level i and every pair of vertices y and y on level j > i, the product of the graphs H x,y and H x,y is isomorphic to the product of the graphs H x,y and H x,y. The details of the product that we use (from [4]) are given below. The notion of isomorphism is the usual (graph-theoretic) one, except that certain short components are dropped, as described below. The precise definition of H x,y is given below, but the reader can think of it as the subgraph between vertex x and vertex y. Some fairly complex graphs H satisfy this condition (see, for example, Figure 5), so for these graphs H the counting problem is in P. Our algorithm for counting graph homomorphisms for tractable digraphs H is based on factoring. A difficulty is that the relevant algebra lacks unique factorisation. We deal with this by introducing preconditioners. See Section 6. Before giving precise definitions and proving our dichotomy theorem, we note that our proof relies on two fundamental results of Bulatov and Grohe [1] and Lovász [10]. These will be introduced below in Section 3. 2 Notation and definitions We use the notation N = {1, 2, 3,...} and N 0 = N {0}. For m, n N 0, we will write [m, n] = {m, m + 1,..., n 1, n} and [n] = [1, n]. We will generally let H = (V, E) denote a fixed colouring digraph, and G = (V, E) an input digraph. We denote the empty digraph (, ) by Homomorphisms Let G = (V, E), H = (V, E). If f : V V, and e = (v, v ) E, we write f(e) = (f(v), f(v )). Then f is a homomorphism from G to H (or an H-colouring of G) if f(e) E. We will denote the number of distinct homomorphisms from G to H by #H(G). Note that #H(0) = 1 for all H. Let f be a homomorphism from H 1 = (V 1, E 1 ) to H 2 = (V 2, E 2 ). If f is also injective, it is a monomorphism. Then E 1 = f(e 1 ) E 2. If there exist monomorphisms f from H 1 to H 2 and f from H 2 to H 1, then f is an isomorphism from H 1 to H 2. Then E 1 = f(e 1 ) E 2 = 2

3 f (E 2 ) E 1, so f(e 1 ) = E 2 which implies f(e 1 ) = E 2. If there is an isomorphism from H 1 to H 2, we write H 1 = H2 and say H 1 is isomorphic to H 2. The relation = is easily seen to be an equivalence. We will usually use H 1 to denote an equivalence class of isomorphic graphs, and write H 1 = H 2 rather than H 1 = H2. In this paper, we consider the particular case where H = (V, E) is a directed acyclic graph (DAG). Thus, in particular, H has no self-loops. Clearly #H(G) = 0 if G is not also a DAG. 2.2 Layered graphs A DAG H = (V, E) is a layered digraph with l layers if V is partitioned into (l + 1) levels V i (i [0, l]) such that (u, u ) E only if u V i 1, u V i for some i [l]. We will allow V i =. We will call V 0 the top and V l the bottom. Nodes in V 0 are called sources and nodes in V l are called sinks. (Note that the usage of the words source and sink varies. In this paper a vertex is called a source only if it is in V 0. A vertex in V i for some i 0 is not called a source, even if it has in-degree 0.) Layer i is the edge set E i E of the subgraph H [i 1,i] induced by V i 1 V i. More generally we will write H [i,j] for the subgraph induced by j k=i V k. Let G l be the class of all layered digraphs with l layers and let C l be the subclass consisting of l-layered digraphs in which every connected component spans all l + 1 levels. If H C l and G = (V, E) C l, with V i denoting level i (i [0, l]) and E i denoting layer i (i [l]), then any homomorphism from G to H is a sequence of functions f i : V i V i (i [0, l]) which induce a mapping from E i into E i (i [l]). We use C l to define an equivalence relation on G l. In particular, for H 1, H 2 G l, H 1 H 2 iff Ĥ1 = Ĥ2, where Ĥi C l is obtained from H i by deleting every connected component that spans fewer than l + 1 levels. 2.3 Sums and products If H 1 = (V 1, E 1 ), H 2 = (V 2, E 2 ) are disjoint digraphs, the union H 1 + H 2 is the digraph H = (V 1 V 2, E 1 E 2 ). Clearly 0 is the additive identity and H 1 + H 2 = H 2 + H 1. If G is connected then #(H 1 + H 2 )(G) = #H 1 (G) + #H 2 (G), and if G = G 1 + G 2 then #H(G) = #H(G 1 )#H(G 2 ). The layered cross product [4] H = H 1 H 2 of layered digraphs H 1 = (V 1, E 1 ), H 2 = (V 2, E 2 ) G l is the layered digraph H = (V, E) G l such that V i = V 1i V 2i (i [0, l]), and ( (u1, u 2 ), (u 1, u 2 )) E if and only if (u 1, u 1 ) E 1 and (u 2, u 2 ) E 2. We will usually write H 1 H 2 simply as H 1 H 2. It is clear that H 1 H 2 is connected only if both H 1 and H 2 are connected. The converse is not necessarily true. See Figure 1. PSfrag replacements = Figure 1: A disconnected product 3

4 Nevertheless, we have the following lemma. Lemma 2.1. If H 1, H 2 C l and both of these graphs contain a directed path from every source to every sink then exactly one component of H 1 H 2 spans all l + 1 levels. The other components do not include level 0 or level l. Proof. There is a directed path from every source of H 1 H 2 to every sink. Thus, the sources and sinks are all in the same connected component. Note that H 1 H 2 = H 2 H 1, using the isomorphism (u 1, u 2 ) (u 2, u 1 ). If G, H 1, H 2 C l then any homomorphism f : G H 1 H 2 can be written as a product f 1 f 2 of homomorphisms f 1 : G H 1 and f 2 : G H 2, and any such product is a homomorphism. Thus #H 1 H 2 (G) = #H 1 (G) #H 2 (G). Observe that the directed path P l of length l gives the multiplicative identity 1 and that 0H = H0 = 0 for all H. It also follows easily that H(H 1 + H 2 ) = HH 1 + HH 2, so distributes over +. The algebra A = (G l, +,, 0, 1) is a commutative semiring. The + operation is clearly cancellative. We will show in Lemma 3.6 that is also cancellative, at least for C l. In many respects, this algebra resembles arithmetic on N 0, but there is an important difference. In A we do not have unique factorisation into primes. A prime is any H G l which has only the trivial factorisation H = 1H. Here we may have H = H 1 H 2 = H 1 H 2 with H 1, H 2, H 1, H 2 prime and no pair equal, even if all the graphs are connected. Example 2.2. Let K 1,m be the usual undirected bipartite clique K 1,m, but with all edges directed from the root vertex, and let K m,1 be K 1,m with all edges reversed. The graphs H 1, H 2, H 1, H 2 will all have three layers. Each has top layer K 1,m1, middle layer a disjoint union of K 1,m s, and bottom layer K m3,1. We specify the subgraphs in each layer in the following table. We show H 1 in Figure 2, where all edges are directed downwards. Figure 2: The graph H 1 in Example 2.2 Graph Layer 1 Layer 2 Layer 3 H 1 K1,4 K1,8 + K 1,2 + K 1,1 + K 1,1 K12,1 H 2 K1,9 K1,8 + K 1,4 + K 1,2 + K 1,1 + K 1,1 + K 1,1 + K 1,1 + K 1,1 + K 1,1 K20,1 H 1 H 2 K 1,6 K1,8 + K 1,4 + K 1,1 + K 1,1 + K 1,1 + K 1,1 K16,1 K 1,6 K1,8 + K 1,2 + K 1,2 + K 1,1 + K 1,1 + K 1,1 K15,1 4

5 It is clear that H 1, H 2, H 1, H 2 are connected, and it is not difficult to show that none of them has a nontrivial factorisation. However, it is easy to verify that H 1 H 2 = H 1 H 2. The layered cross product was defined in [4] in the context of interconnection networks. It is similar to the (non-layered) direct product [9], which also lacks unique factorisation, but they are not identical. In general, they have different numbers of vertices and edges. 3 Fundamentals Our proof relies on two fundamental results of Bulatov and Grohe [1] and Lovász [10]. First we give the basic result of Lovász [10]. (See also [8, Theorem 2.11].) If H = (V, E), G = (V, E) are DAGs, we denote the number of monomorphisms from G to H by H(G). The following is essentially a special case of Lovász [10, Theorem 3.6], though stated rather differently. We give a proof since it yields additional information. Theorem 3.1 (Lovász). If #H 1 (G) = #H 2 (G) for all G, then H 1 = H 2. Proof. Let f be any homomorphism from G to H. Then f 1 induces a partition I of V into disjoint sets S I,1,..., S I,kI such that each S I,i (i [k I ]) is independent in G. Each partition I fixes subsets S I,i V which map to the same u i V. Let I be the set of all such partitions. With the relation I I whenever I is a refinement of I, P G = (I, ) is a poset. Note that P G depends only on G. We will write for the partition into singletons, so I for any I I. Let G/I be the digraph obtained from G by identifying all vertices in S I,i (i [k I ]). Then we have #H(G) = #H(G/ ) = I I H(G/I) = I I H(G/I)ζ(, I), where ζ(i, I ) = 1 if I I, and ζ(i, I ) = 0 otherwise, defines the ζ-function of P G. More generally, the same reasoning gives #H(G/I) = I I H(G/I ) = I I H(G/I )ζ(i, I ). Now Möbius inversion for posets [11, Ch. 25] implies that the matrix ζ : I I {0, 1} has a unique inverse µ : I I Z. It follows directly that H(G) = I I #H(G/I)µ(, I). Using the assumption of the theorem, for every G we now have H 1 (G) = I I #H 1 (G/I)µ(, I) = I I #H 2 (G/I)µ(, I) = H 2 (G). (1) In particular this implies H 2 (H 1 ) = H 1 (H 1 ) > 0, so there is a monomorphism f from H 1 into H 2. By symmetry, there is also a monomorphism f from H 2 into H 1. Remark 3.2. In this paper, the digraph H is always a DAG, so it has no self-loops. However, if we generalise to the situation in which H 1 and H 2 can have both looped and unlooped vertices (but every vertex in G is unlooped), as in the usual definition of the general H- colouring problem [3], the above proof is no longer valid. The reason is that S I,i must be 5

6 an independent set if u i is unlooped, but can be arbitrary if u i is looped. Thus P G no longer depends only on G. However, if H 1 and H 2 have all their vertices looped, an obvious modification of the proof goes through. Whether the theorem remains true for H with both looped and unlooped vertices is, as far as we know, an open question. Note that, if G C l, the poset P G is a lattice since it possesses a (unique) element such I for all I I, with G/ = P l. The following variant of Theorem 3.1 is useful. Theorem 3.3. If H 1, H 2 C l and #H 1 (G) = #H 2 (G) for all G C l, then H 1 = H 2. The proof follows from the proof of Theorem 3.1 since the only instances of G used in that proof are of the form H 1 /I or H 2 /I and these are in C l. Similar reasoning gives the following corollaries. Corollary 3.4. Suppose H k = (V k, E k ) (k = 1, 2). If there is any G with #H 1 (G) #H 2 (G), there is a G such that 0 < V max k=1,2 V k. Proof. Clearly G must be nonempty. Assume V 1 V 2. If V 1 < V 2, taking G 0 = ({v 0 }, ) gives #H 1 (G 0 ) = V 1 V 2 = #H 2 (G 0 ), so we may take V = 1 V 2. Otherwise, since H 1 H 2, either H 2 (H 1 ) H 1 (H 1 ) or H 2 (H 2 ) H 1 (H 2 ). In the former case, we can use Equation 1 with G = H 1 to see that one of the H 1 /I must be such that #H 1 (H 1 /I) #H 2 (H 1 /I). In the latter case, one of the H 2 /I must be such that #H 2 (H 2 /I) #H 1 (H 2 /I). But all the graphs H 1 /I, H 2 /I have at most max i=1,2 V i vertices. Corollary 3.5. Suppose H 1 = (V 1, E 1 ), H 2 = (V 2, E 2 ) C l. If there is any G C l with #H 1 (G) #H 2 (G), there is such a G such that 0 < V max k=1,2 V k. Therefore we can find a witness to the predicate G : #H 1 (G) #H 2 (G), if one exists, among the graphs of the form H 1 /I and H 2 /I. Lemma 3.6. If H 1 H = H 2 H for H 1, H 2, H C l, then H 1 = H 2. Proof. Suppose G C l. Since H C l, #H(G) 0. Also, since H 1 and H 2 are in C l and H 1 H = H 2 H, #H 1 (G)#H(G) = #H 2 (G)#H(G) so #H 1 (G) = #H 2 (G). Now use Theorem 3.3. Here is another similar lemma that we will need. Lemma 3.7. If H 1, H 2, H C l and each of these contains a directed path from every source to every sink and H 1 H H 2 H then H 1 = H 2. Proof. Suppose G C l. Since H C l, #H(G) 0. Let Ĥ1 be the single full component of H 1 H from Lemma 2.1. Similarly, let Ĥ2 be the single full component of H 2 H. Then since G C l, #Ĥ1(G) = #H 1 H(G) = #H 1 (G)#H(G) and #Ĥ2(G) = #H 2 H(G) = #H 2 (G)#H(G). So since #Ĥ1(G) = #Ĥ2(G) by the equivalence in the statement of the lemma, we have #H 1 (G) = #H 2 (G). Now use Theorem

7 The second fundamental result is a theorem of Bulatov and Grohe [1, Theorem 1], which provides a powerful generalisation of a theorem of Dyer and Greenhill [3]. Let A = (A ij ) be a k k matrix of non-negative rationals. We view A as a weighted digraph such that there is an edge {i, j} with weight A ij if A ij > 0. Given a digraph G = (V, E), Eval(A) is the problem of computing the partition function Z A (G) = A σ(u)σ(v). (2) σ:v {1,...,k} (u,v) E In particular, if A is the adjacency matrix of a digraph H, Z A (G) = #H(G). Thus Eval(A) has the same complexity as #H. If A is symmetric, corresponding to a weighted undirected graph, the following theorem characterises the complexity of Eval(A). Theorem 3.8 (Bulatov and Grohe). Let A be a non-negative rational symmetric matrix. (1) If A is connected and not bipartite, then Eval(A) is in polynomial time if the row rank of A is at most 1; otherwise Eval(A) is #P-complete. (2) If A is connected and bipartite, then Eval(A) is in polynomial time if the row rank of A is at most 2; otherwise Eval(A) is #P-complete. (3) If A is not connected, then Eval(A) is in polynomial time if each of its connected components satisfies the corresponding condition stated in (1) or (2); otherwise Eval(A) is #P-complete. 4 Reduction from acyclic H to layered H Let H = (V, E) be a DAG. Clearly #H is in #P. We will call H easy if #H is in P and hard if #H is #P-complete. We will show that H is hard unless it can be represented as a layered digraph. Let N k (u, u ) be the number of paths of length k from u to u in H. Say that vertices u, u V are k-compatible if, for some vertex w, there is a length-k path from u to w and from u to w. We say that H is k-good if, for every k-compatible pair (u, u ), there is a rational number λ such that N k (u, w) = λn k (u, w) ( w V). Lemma 4.1. If there is a k such that H is not k-good then #H is #P-complete. Proof. Fix k, u, and u such that u and u are k-compatible, but there is no λ. Let A be the adjacency matrix of H and let A = A k (A k ) T. Note that (A k ) u,w = N k (u, w). First, we show that Eval(A ) is #P-hard. Note that A is symmetric with non-negative rational entries. Let A be the square sub-matrix of A corresponding to the connected component containing u and u. Note that u and u are in the same connected component since they are k-compatible. Also, there are loops on vertices u and u, so A is not bipartite. To show that Eval(A ) is #P-complete, we need only show that the rank of A is bigger than 1. To do this, we just need to find a 2 2 submatrix that is non-singular, i.e. with nonzero determinant. Take the principal submatrix indexed by rows u and u and columns u and u. The determinant is ( w N k(u, w) 2) ( w N k(u, w) 2) ( w N k(u, w)n k (u, w)) 2. By Cauchy-Schwartz, the determinant is non-negative, and is zero only if λ exists, which we have assumed not to be the case. Thus Eval(A ) is #P-complete. 7

8 Now we use the hardness of Eval(A ) to show that Eval(A) is #P-hard. To do this, take an undirected graph G which is an instance of Eval(A ). Construct a digraph G by taking every edge {v, v } of G and replacing it with a digraph consisting of a directed length-k path P k from v to a new vertex w and a directed length-k path P k from v to w. Note that Eval(A) on G is the same as Eval(A ) on G. Thus Eval(A) is #P-complete, and #H has the same complexity. Remark 4.2. We have used the path P k as a gadget in the above reduction, in order to simulate an edge of an undirected graph. We can use any other DAG G having a single source and single sink in the same way, and we do that below. Remark 4.3. The statement of Lemma 4.1 is not symmetrical with respect to the direction of edges in H. However, let us define the digraph H R to be that obtained from H by reversing every edge. Then #H R and #H have the same complexity. To see this, simply observe that #H R (G R ) = #H(G) for all G. We are now in a position to prove the main result of this section. Lemma 4.4. If H is a DAG, but it cannot be represented as a layered digraph, then #H is #P-hard. Proof. Suppose that H contains two paths of different lengths from u to u. Choose k > 0, the length of the short path as small as possible. Choose k > k, the length of the long path, as large as possible subject to the choice of k. Suppose that k edges of the long path take us from u to b and that k edges of the long path take us from a to u. We claim that H has no length-k path from a to b. Since u and a are k-compatible, Lemma 4.1 will then give the conclusion. If k 2k, the claim follows from the fact that H has no cycles. (Either a = b or there is a path from b to a on the long path.) If k < 2k then the claim follows from the choice of k. If there were a length-k path from a to b then we could go from u to a following the long path, from a to b on a k-edge path and from b to u again following the long path. The concatenation of these paths would have length greater than k. 5 A structural condition for hardness We now formulate a sufficient condition for hardness of a layered digraph H = (V, E) G l. Suppose s V i and t V j for i < j. If there is a directed path in H from s to t, we let H st be the subgraph of H induced by s, t, and all components of H [i+1,j 1] to which both s and t are incident. Otherwise, we let H st = 0. Lemma 5.1. If there exist x, x V 0, y, y V l such that H xy H x y H xy H x y, and at most one of H xy, H xy, H x y, H x y is 0, then H is #P-complete. Proof. If exactly one of H xy, H xy, H x y, H x y is 0 then Lemma 4.1 applies. Suppose that none of H xy, H xy, H x y, H x y is 0. Note that x, x, y and y are all in the same component of H and that this component is in C l. By Lemma 2.1, H xy H x y contains a single component H xy;x y that spans all l + 1 levels. So if G C l, #H xy H x y (G) = #H xy;x y (G). Similarly, H xy H x y contains a single component H xy ;x y that spans all l+1 levels and #H xy H x y(g) = #H xy ;x y(g). By the assumption in the lemma, H xy;x y H xy ;x y. So, by Theorem 3.3, there is a Γ C l such that #H xy;x y (Γ) #H xy ;x y(γ). (3) 8

9 We can assume without loss of generality that Γ has a single source and sink since H xy;x y and H xy ;x y do. By Corollary 3.5, we can also ensure that Γ has at most V vertices. Equation (3) implies #H xy (Γ)#H x y (Γ) #H xy (Γ)#H x y(γ). (4) We now follow the proof of Lemma 4.1, replacing the path gadget with Γ as indicated in Remark 4.2. The matrix A is indexed by sources u of H and sinks w. A uw is #H uw (Γ). Then A = AA T so A uu = w A uwa u w. As in the proof of Lemma 4.1, we first show that Eval(A ) is #P-hard. Let A be the square submatrix corresponding to the connected component containing x and x. Note that they are in the same connected component since Γ C l and H x,y 0 and H x y 0 so A xy 0 and A x y 0. Consider the principal submatrix indexed by rows x and x and columns x and x. The determinant is ( ) ( ) ( 2 A xw A x w). w A 2 xw w A 2 x w As before, this is zero only if there is a λ such that A xw = λa x w for all w and this is false by (4) which says A xy A x y A xy A x y. Thus, the rank of A is bigger than 1 and Eval(A ) is #P-complete. Now let B be the adjacency matrix of H. Reduce Eval(A ) to Eval(B) as follows. Take an undirected graph G which is an instance of Eval(A ). Construct a digraph G by taking every edge {v, v } of G and replacing it with a digraph consisting of a copy of Γ with source v and a copy of Γ with source v with the sinks identified. Note that Eval(B) on G is the same as Eval(A ) on G. Thus Eval(B) is #P-complete, and #H has the same complexity. Clearly checking the condition of the Lemma and carrying out the search for the gadget Γ both require only constant time. Note that if x, x, y, y are not all in the same component of H then at least two of H xy, H x y, H xy, H x y are 0, so Lemma 5.1 has no content. w x x x x PSfrag replacements y y H y H xy y H x y Figure 3: The graph of Example 5.2 Example 5.2. Consider the H in Figure 3. We have H xy H x y = H xy and H xy H x y = H x y. Clearly H xy and H x y are not isomorphic, so this H is #P-complete. A suitable gadget is 9

10 Γ = H xy. The following table gives #H xy (Γ), #H xy (Γ), #H x y(γ) and #H x y (Γ). Since this matrix has rank 2 and is indecomposable, we can prove #P-completeness using Γ as a gadget. y y x 4 1 x 2 1 We may generalise Lemma 5.1 as follows. Lemma 5.3. If there exist x, x V i, y, y V j ( 0 i < j l) such that H xy H x y H xy H x y, and at most one of H xy, H x y, H xy, H x y is 0, then H is #P-complete. Proof. If l = j i, consider any G having l layers. For k = 0, 1,..., l l, let H k be the subgraph of H induced by V k V k+1 V k+l. Then #H(G) = #H k (G), l l k=0 so colouring with H is equivalent to colouring with the graph H = (V, E ) = l l k=0 H k G l. But x, x are in V 0, and c, d in V l. The result now follows from Lemma 5.1. Note that the N of Bulatov and Dalmau [2] is the special case of Lemma 5.3 in which j = i + 1, H xy = H xy = H x y = 1, and H x y = 0. More generally, any structure with H xy, H xy, H x y 0 and H x y = 0 is a special case of Lemma 5.3, so is sufficient to prove #P-completeness. Such a structure is equivalent to the existence of paths from x to y, x to y and x to y, when no path from x to y exists. We will call such a structure an N, since it generalises Bulatov and Dalmau s [2] construction. Lemma 5.4. If H C l is connected and not hard, then there exists a directed path from every source to every sink. Proof. Clearly, by Lemma 5.3, H must be N-free. Since it is connected, there is an undirected path from any x V 0 to any y V l. Suppose that the shortest undirected path from x to y is not a directed path. Then some part of it induces an N in H, giving a contradiction. Lemma 5.4 cannot be generalised by replacing source with node (at any level) with indegree 0 and replacing sink similarly, as the graph in Figure 4 illustrates. We will call four vertices x, x, y, y in H, with x, x V i and y, y V j (0 i < j l), a Lovász violation 2 if at most one of H xy, H x y, H xy, H x y is 0 and H xy H x y H xy H x y. A graph H with no Lovász violation will be called Lovász-good. We show next that this property is preserved under the layered cross product. Lemma 5.5. If H, H 1, H 2 C l and H = H 1 H 2 then H is Lovász-good if and only if both H 1 and H 2 are Lovász-good. Proof. Let H k = (V k, E k ) with levels V k,i (k = 1, 2; i [0, l]) and H = (V, E) with levels V i (i [0, l]). Suppose x k, x k V k,i, y k, y k V k,j (k = 1, 2; 0 i < j l), not necessarily 2 The name derives from the isomorphism theorem (Theorem 3.1) of Lovász. 10

11 x s PSfrag replacements t y Figure 4: An easy H with no st path distinct. We will write, for example, x 1 x 2 V i for product vertices and H 1 x,y for (H 1 ) x,y. Then H x1 x 2,y 1 y 2 H x 1 x 2,y 1 y 2 = H 1 x 1,y 1 H 2 x 2,y 2 H 1 x 1,y 1 H2 x 2,y 2 and H x1 x 2,y 1 y 2 H x 1 x 2,y 1y 2 = H 1 x 1,y 1 H2 x 2,y 2 H1 x 1,y 1 H2 x 2,y 2 (5) (6) Now, if Hx 1 1,y 1 Hx 1 H 1 1,y 1 x 1,y H 1 1 x 1,y and H2 1 x 2,y 2 Hx 2 H 2 2,y 2 x 2,y 2H 2 x 2,y, then (5) and (6) imply 2 H x1 x 2,y 1 y 2 H x 1 x H 2,y 1 y 2 x 1 x 2,y 1 H y 2 x 1 x 2,y 1y 2. This if H 1 and H 2 are Lovász-good, so is H. Conversely, suppose without loss of generality that H 1 is not Lovász-good, and that H 1 x 1,y 1 H 1 x 1,y 1 H1 x 1,y 1 H1 x 1,y 1. Taking x 2 = x 2, y 2 = y 2 for any vertices x 2 V 2,i, y 2 V 2,j such that H 2 x 2,y 2 0, from (5) and (6) we have H x1 x 2,y 1 y 2 H x 1 x 2,y 1 y 2 = H1 x 1,y 1 H 1 x 1,y 1 (H2 x 2,y 2 ) 2 and H x1 x 2,y 1 y 2 H x 1 x 2,y 1 y 2 = H1 x 1,y 1 H1 x 1,y 1 (H2 x 2,y 2 ) 2. First, suppose that none of Hx 1 1,y 1, Hx 1, H 1 1,y 1 x 1,y H 1, 1 x 1,y is 0. Let Z 1 be the full component of 1 Hx 1 1,y 1 Hx 1 according to Lemma 2.1. Similarly, let Z 2 be the full component of H 1 1,y 1 x 1,y 1H 1 x 1,y. 1 Let Z = (Hx 2 2,y 2 ) 2. To show that H is not Lovász-good, we wish to show Z 1 Z Z 2 Z. By Lemma 3.7, this follows from Z 1 Z 2. Finally, suppose that exactly one of Hx 1 1,y 1, Hx 1, H 1 1,y 1 x 1,y H 1, 1 x 1,y is 0. Then exactly one of 1 H x1 x 2,y 1 y 2 H x 1 x 2,y 1 y and H 2 x 1 x 2,y 1 y 2 H x 1 x 2,y 1 y is 0, so they are not equivalent under and H 2 is not Lovász-good. The requirement of H being Lovász-good is essentially a rank 1 condition in the algebra A of Section 2.3, and therefore resembles the conditions of [1, 3]. However, since A lacks unique factorisation, difficulties arise which are not present in the analyses of [1, 3]. But a more important difference is that, whereas the conditions of [1, 3] permit only trivial easy graphs, Lovász-good graphs can have complex structure. See Figure 5 for a small example. 11

12 Figure 5: A Lovász-good H 6 Main theorem We can now state the dichotomy theorem for counting homomorphisms to directed acyclic graphs. Theorem 6.1. Let H be a directed acyclic graph. Then #H is in P if H is layered and Lovász-good. Otherwise #H is #P-complete. The proof of Theorem 6.1 will use the following lemma, which we prove later. Lemma 6.2. Suppose H C l is connected, with a single source and sink, and is Lovász-good. There is a polynomial-time algorithm for the following problem. Given a connected G C l with a single source and sink, compute #H(G). Proof of Theorem 6.1. We have already shown that any non-layered H is hard in Lemma 4.4. We have also shown in Lemma 5.3 that H is hard if it is not Lovász-good. Suppose H G l is Lovász-good. We will show how to compute #H(G). First, we may assume that G is connected since, as noted in Section 2.3, if G = G 1 + G 2 then #H(G) = #H(G 1 )#H(G 2 ). We can also assume that H is connected since, for connected G, #(H 1 + H 2 )(G) = #H 1 (G) + #H 2 (G), but H 1 and H 2 are Lovász-good if H 1 + H 2 is. So we can now assume that H C l is connected and G is connected. If G has more than l + 1 non-empty levels then #H(G) = 0. If G has fewer than l non-empty levels then decompose H as in the Proof of Lemma 5.3. So we can assume without loss of generality that both H and G are connected and in C l. Now we just add a new level at the top of H with a single vertex, adjacent to all sources of H and a new level at the bottom of H with a single sink, adjacent to all sinks of H. We do the same to G. Then we use Lemma 6.2. Before proving Lemma 6.2 we need some definitions. Suppose H is a connected graph in C l. For a subset S of sources of H, let H [0,j] S be the subgraph of H [0,j] induced by those vertices from which there is an (undirected) path to S in H [0,j]. We say that H is top-j disjoint if, for every pair of distinct sources s, s, H [0,j] {s} and H[0,j] {s } are disjoint. We say that H is bottom-j disjoint if the reversed graph H R from Remark 4.3 is top-j disjoint. Finally, We say that H is fully disjoint if it is top-(l 1) disjoint and bottom-(l 1) disjoint. We will say that (Q, U, D) is a good factorisation of H if Q, U and D are connected Lovászgood graphs in C l such that QH UD, 12

13 Q has a single source and sink, U has a single sink, and D has a single source. Remark 6.3. The presence of the preconditioner Q in the definition of a good factorisation is due to the absence of unique factorisation in the algebra A. Our algorithm for computing homomorphisms to an Lovász-good H works by factorisation. However, it is possible to have a non-trivial Lovász-good H which is prime. A simple example can be constructed from the graphs H 1, H 2, H 1, H 2 of Example 2.2, by identifying the sources in H 1, H 1 and in H 2, H 2, and the sinks in H 1, H 2 and in H 1, H 2. The resulting 2-source, 2-sink graph has no nontrivial factorisation. We use the following operations on a Lovász-good connected digraph H C l. Local Multiplication: Suppose that U is a connected Lovász-good single-sink graph in C j on levels 0,..., j for j l. Let C be a Lovász-good connected component in H [0,j] with no empty levels. Then Mul(H, C, U) is the graph constructed from H by replacing C with the full component of UC. (Note that there is only one full component, by Lemma 5.4 and 2.1.) Local Division: Suppose S V 0, and that (Q, U, D) is a good factorisation of H [0,j] S. Then Div(H, Q, U, D) is the graph constructed from H by replacing H [0,j] S with D. We can now state our main structural Lemma. Lemma 6.4. If H C l is connected, and Lovász-good, then it has a good factorisation (Q, U, D). We prove Lemma 6.4 below in Section 7. In the course of the proof, we give an algorithm for constructing (Q, U, D). We now describe how we use Lemma 6.4 (and the algorithm) to prove Lemma 6.2. Proof of Lemma 6.2. The proof is by induction on l. The base case is l = 2. (Note that calculating #H(G) is easy in this case.) For the inductive step, suppose l > 2. Let H denote the part of H excluding levels 0 and l and let G denote the part of G excluding levels 0 and l. Using reasoning similar to that in the proof of Theorem 6.1, we can assume that G is connected and then that H is connected. Since H is Lovász-good, so is H. Now by Lemma 6.4 there is a good factorisation (Q, U, D ) of H. Let S V 1 be the nodes in level 1 of H that are adjacent to the source and T V l 1 be the nodes in level l 1 of H that are adjacent to the sink. Note that V 1 is the top level of U and V l 1 is the bottom level of D. Construct Q from Q by adding a new top and bottom level with a new source and sink. Connect the new source and sink to the old ones. Construct D from D by adding a new top and bottom level with a new source and sink. Connect the new source to the old one and the new sink to T. Finally, construct U from U by adding a new top and bottom level with a new source and sink. Connect the new source to S and the new sink to the old one. See Figure 6. Note that (Q, U, D) is a good factorisation of H. To see that QH UD, consider the component of Q H that includes sources and sinks. (There is just one of these. Since H is Lovász-good, it has a directed path from every source to every sink by Lemma 5.4. So does Q. Then use Lemma 2.1.) This is isomorphic to the corresponding component in D U since (Q, U, D ) is a good factorisation of H. The isomorphism maps S in H to a corresponding S in U and now note that the new top level is appropriate in QH and DU. Similarly, the new bottom level is appropriate. Now let s consider how to compute #Q(G). In any homomorphism from G to Q, every node in level 1 of G gets mapped to the singleton in level 1 of Q. Thus, we can collapse all level 1 13

14 PSfrag replacements S S Q H D U T T Figure 6: H nodes of G into a single vertex without changing the problem. At this point, the top level of G and Q are not doing anything, so they can be removed, and we have a sub-problem with fewer levels. So #Q(G) can be computed recursively. The same is true for #D(G) and #U(G). Since G C l, #QH(G) = #Q(G)#H(G). Also, since components without sources and sinks cannot be used to colour G (which has the full l layers), this is equal to #D(G)#U(G). Thus, we can output #H(G) = #D(G)#U(G)/#Q(G). That concludes the proof of Theorem 6.2, so it only remains to prove Lemma 6.4. The proof will be by induction, and we will need the following technical lemmas in the induction. Lemma 6.5. Suppose that H is Lovász-good, top-(j 1) disjoint, S V 0 and H [0,j] S is connected. If (Q, U, D) is a good factorisation of H [0,j] S, then Div(H, Q, U, D) is Lovász-good. Proof. Suppose to the contrary that H = Div(H, Q, U, D) contains a Lovász violation x, x, y, y, with x, x in level i, and y, y in level k. Since D is Lovász-good, we may assume without loss that x is in D, with i < j, and that k > j. Thus y, y are not in D. Let Q, Ũ be Q[i,j], U [i,j], both extended downwards by a single path to level k. There are two cases. If x is in D, Lemma 5.5 gives us a Lovász violation in HŨ [i,k]. (By the proof of Lemma 5.5, this involves nodes at level i and k.) This gives us a Lovász violation in H [i,k] Q[i,k] since these graphs are isomorphic except for components that don t extend down to y and y. Finally, by Lemma 5.5, we find a Lovasz violation in H [i,k], and hence in H, which gives a contradiction. If x is not in D, then let H be Mul( H, D, U). Let w be some vertex on level i of Ũ which has a directed path to the sink z of Ũ. Now by construction H xw,yh x,y H xy Hx y (Ũwz), H xw,y H x,y H xy Hx y(ũwz). Because these graphs all have single sources and sinks (at levels i and k), we can apply Lemma 3.7 to cancel and find a Lovász violation xw, y, x, y in H. Then do a local multiplication multiplying the component of x by Q and again, in the same way, we find a Lov asz violation xw, y, x w, y in the resulting graph, H 14

15 Now since DU QH [0,j] S and since xw does have a path down to level j of H, the component containing xw in H [0,j] is isomorphic to the corresponding component in QH [0,j]. Thus, the same Lovász violation exists in QH (where now we extend the tail of Q all the way down to level l of H. Now to get a Lovász violation in H itself, we use the reasoning from the proof of Theorem 5.5. Say the Loász violation is x Q x H, y Q y H, x Q x H, y Q y H. Then [i,k] [i,k] ( QH) x Q x H y Q y H ( QH) Q x Q x H y Q y xq y Q H xh y H Q x H Q y Q H x, H y H [i,k] [i,k] ( QH) x Q x H ( QH) y Q y H x Q x H,y Qy H Q xq y Q H x H y H Q x Q y H Q x H y. H And since Q is Lovász-good, the first of these is equivalent to Q xq y Q H x H y H Q x Q y Q H x H y H. Now if we had H xh y H H x H y H H x H y H H x H y H we could multiply both sides by Q xq y Q Q x Q y Q to get [i,k] [i,k] [i,k] [i,k] ( QH) x Q x H y Q y H ( QH) ( QH) x Q x H y Q y H x Q x H ( QH) y Q y H x Q x H,y Qy H which is a contradiction since this was a Lovász violation. Lemma 6.6. Suppose that H and U are Lovász-good. Then Mul(H, C, U) is Lovász-good. Proof. Suppose M = Mul(H, C, U) is not Lovász-good. By Lemma 5.5, the full component of UC is Lovász-good, so there is a Lovász violation in levels i and k of M with i j and k > j. Also, one of the relevant vertices in level i is in the component from UC in the local multiplication. First, suppose that both of the relevant vertices in level i are in this component. Then there is a Lovász violation (the same one) in ÛH, where Û extends U with a path. Now restricting attention to levels i k, Lemma 5.5 shows that there is a Lovász violation in H or Û (hence U), which is a contradiction. Second, suppose that only one of the relevant vertices in level i is in the component from CU in the local multiplication. Then the Lovász violation can be written M [i,k] U xc x,y M [i,k] x,y M [i,k] U xc x,y M [i,k] x,y and expanding the product, letting node s on level k, this is Û denote the extension of U downwards with a path to Û [i,k] U x,s H[i,k] C x,y H[i,k] x,y Û [i,k] U x,s H[i,k] C x,y H [i,k] x,y. But this gives us a Lovász violation in H, which is a contradiction. 15

16 7 Proof of Lemma 6.4 A top-dangler is a component in H [1,l 1] that is incident to a source but not to a sink. Similarly, a bottom-dangler is a component in H [1,l 1] that is incident to a sink but not to a source. (Note that a bottom-dangler in H is a top-dangler in H R.) The proof of Lemma 6.4 will be by induction. The base case will be l = 1, where it is easy to see that a connected Lovász-good H must be a complete bipartite graph. The ordering for the induction will be lexicographic on the following criteria (in order). 1. the number of levels, 2. the number of sources, 3. the number of top-danglers, 4. the number of sinks, 5. the number of bottom-danglers. Thus, for example, if H has fewer levels than H then H precedes H in the induction. If H and H have the same number of levels, the same number of sources and the same number of top-danglers but H has fewer sinks then H precedes H in the induction. The inductive step will be broken into cases. The cases are exhaustive but not mutually exclusive given an H we will apply the first case that applies. Case 1: H is top-(j 1) disjoint and has a top dangler with depth at most j 1. Case 2: For j < l, H is top-(j 1) disjoint, but not top-j disjoint, and has no top dangler with depth at most j 1. Case 3: H is top-(l 1) disjoint and has no top dangler and is bottom-(j 1) disjoint and has a bottom dangler with height at most (j 1). Case 4: For j < l, H is top-(l 1) disjoint and has no top dangler and is bottom-(j 1) disjoint, but not bottom-j disjoint, and has no bottom dangler with height at most (j 1). Case 5: H is fully disjoint. and has no top danglers or bottom danglers. 7.1 Case 1 Let R be a top dangler with depth at most j 1 (meaning that it is a component in H [1,...,l 1] that is incident to a source but not to a sink, also that levels j,..., l 1 are empty). Note that since H is top-(j 1) disjoint, R must be adjacent to a single source, v, in H. This follows from the definition of top-(j 1) disjoint, and from the fact that R has depth at most j 1. Define j to be the highest level on which R is non-empty. Note that j {1,..., j 1}. Construct H from H by removing R. Note that H is connected. By construction (from H), H is Lovász-good, and has no non-empty levels. It precedes H in the induction order since it has the same number of levels, the same number of sources and one fewer top-dangler. By induction, it has a good factorisation (Q, U, D ) so Q H U D. 16

17 Construct D as follows. On layers 1 j, D is identical to D. On layers j + 2 l, D is a path. Every node in level j is connected to the singleton vertex in level j + 1. Then clearly DQ H ÛD where Û is the single full component of DU. (There is just one of these. Since D is Lovász-good, it has a directed path from every source to every sink by Lemma 5.4. So does U. Then use Lemma 2.1.) Note that Û has a single sink. Let R be the graph obtained from R by adding the source v. Let R be Q [0,j ] R. Form U from R and Û by identifying v with the appropriate source of Û. (Note that Û has the same sources as H). Then DQ H U D. Thus, we have a good factorisation (Q, U, D ) of H by taking Q to be the full component of DQ and U to be the full component of U. To see that it is a good factorisation, use Lemma 5.5 to show that Q and U are Lovász-good. 7.2 Case 2 Partition the sources of H into equivalence classes S 1,..., S k so that the graphs H [0,j] S i are connected and pairwise disjoint. See Figure 7. Since H is not top-j disjoint some equivalence class, say S 1, contains more than one source. V 0 S 1 S 2 S k Ĥ V j PSfrag replacements Figure 7: H Let Ĥ denote H[0,j] S 1. Ĥ is shorter than H, so it comes before H in the induction order. It is connected by construction of the equivalence classes, and it is Lovász-good by virtue of being a subgraph of H. By induction we can construct a good factorisation ( Q, Û, D) of Ĥ. Let H = Div(H, Q, Û, D). See Figure 8. H comes before H in the induction order because it has the same number of levels, but fewer sources. To see that H is connected note that H is connected and Ĥ is connected. Since ( Q, Û, D) is a good factorisation), we know D is connected, so H is connected. By Lemma 6.5, H is Lovász-good. By induction, we can construct a good factorisation (Q, U, D ) of H. Let s be the (single) source of D. By construction, the sources of U are {s} S 2 S k. See Figure 9. Let C 1,..., C z be the connected components of U [0,j]. Let C 1 be the component containing s. Since the H [0,j] S i are connected and pairwise disjoint, and Q has a single source, 17

18 V 0 s S 2 S k D V j PSfrag replacements Figure 8: H there is a single connected component of (Q H ) [0,j] containing all of S i (and no other sources) so (see Figure 9) there is a single connected component of U [0,j] containing all of S i (and no other sources). For convenience, call this C i. If z > k then components C k+1,..., C z do not contain any sources. (They are due to danglers in U, which in this case are nodes that are not descendants of a source.) Now consider U 1 = Mul(U, C 1, Û). U 1 is the graph constructed from U by replacing C 1 with the full component of C 1 Û. For i {2,..., z}, let U i = Mul(U i 1, C i, Q). U z is the graph constructed from U by replacing C 1 with the full component of C 1 Û and replacing every other C i with the full component of C i Q. Let Q extend Q down to level l with a single path. We claim that Q QH Uz D. (7) To establish Equation (7) note that on levels [j,..., l] the left-hand side is (Q QH) [j,l] (Q H ) [j,l]. Any components of Q H that differ from U D do not include level l, so this is equivalent to (U D ) [j,l] (U z D ) [j,l], which is the right-hand side. So focus on levels 0,..., j. From the left-hand side, look at the component (Q [0,j] QH) S 1. Note that it is connected. It is (Q [0,j] QĤ) [0,j] S 1 which is the right-hand side. (Q [0,j] Û Then look at the component S 2. (Q QH) [0,j] S 2 D) [0,j] S 1 ((D [0,j] U [0,j] ){s} Û) [0,j] (Q [0,j] QH [0,j] ) S2 ((D [0,j] U [0,j] ) S2 Q) [0,j] S 2 18 S 1 (D U z ) [0,j] S 1, (D U z ) [0,j] S 2.

19 s S 2 S k s S 2 S k PSfrag replacements Q H D U D Figure 9: Q H D U The other components containing sources (which are the only components that we care about) are similar. Having established (7), we observe that (Q, U, D ) is a good factorisation of H where Q is the full component of Q Q and U is the full component of Uz. Use Lemma 5.5 to show Q is Lovász-good and Lemma 6.6 to show that U z is. 7.3 Case 3 We apply the analysis of Section 7.1 to the reversed graph from Remark 4.3. Let H R be an instance in Case 3. Thus H is top-(j 1) disjoint and has a top-dangler with depth at most j 1. Also, H is bottom-(l 1) disjoint and has no bottom dangler. Apply the transformation in Section 7.1 to H. This produces as inductive instance H. As we noted in Section 7.1, H precedes H in the induction order since it has the same number of levels, the same number of sources and one fewer top-dangler. Crucially, H has the same number of sinks as H and the same number of bottom-danglers as H. (We have not added any.) Thus, H R precedes H R in the inductive order. It has one fewer bottom-dangler and everything else is the same. Then the good factoring (Q, U, D) of H that we produce gives us a good factoring (Q R, D R, U R ) of H R. 7.4 Case 4 We apply the analysis of Section 7.2 to the reversed graph from Remark 4.3. Let H R be an instance in Case 4. The reversed graph H is bottom-(l 1)-disjoint with no bottom dangler. For some j < l, it is top-(j 1)-disjoint, but not top-j disjoint and has no top-dangler with depth at most j 1. Apply the analysis in Case 3. This produces a recursive instance H with fewer sources. H has the same number of levels as H. Furthermore, H has the same number of sinks as H and the same number of bottom danglers. Thus, H R has fewer sinks than H R, but the same number of levels, sources, and top-danglers. So it precedes H R in the induction order. Then the good factoring (Q, U, D) of H that we produce gives us a good factoring (Q R, D R, U R ) of H R. 19

20 7.5 Case 5 In the fully disjoint case, the subgraphs H st (s V 0, t V l ) satisfy {s}, if s = s, t t ; H st H s t = {t}, if s s, t = t ;, (8), if s s, t t. H st 0 and H st H s t H st H s t, since H is Lovász-good. We will assume without loss that V 0 > 1 and V l > 1, since otherwise (1, H, 1) or (1, 1, H) is a good factorisation. See Figure 10. V 0 s 1 s 2 s 3 PSfrag replacements H st V l t 1 t 2 t 3 Figure 10: Fully disjoint case Choose any s V 0, t V l, and let Q = H s t. Note that Q is connected with a single source and sink, and is Lovász-good because H is Lovász-good. Let D be the subgraph t V l H s t of H, and let U be the subgraph s V 0 H st of H. These are both connected and Lovász-good since H is. Clearly D has a single source and U has a single sink. Also QH DU follows from (8) and from the fact that there are no top-danglers or bottom-danglers and (DU) s s,tt = D s tu st H s th st = H s t H st = QH st (s V 0 ; t V l ), (9) where we have used the fact that H is Lovász-good. Thus (Q, U, D) is a good factorisation of H. References [1] A. Bulatov and M. Grohe, The complexity of partition functions, in Automata, Languages & Programming: 31st International Colloquium, Lecture Notes in Computer Science 3142, pp , [2] A. Bulatov and V. Dalmau, Towards a dichotomy theorem for the counting constraint satisfaction problem, in Proc. 44th IEEE Symposium on Foundations of Computer Science, IEEE, pp ,

21 [3] M. Dyer and C. Greenhill, The complexity of counting graph homomorphisms, Random Structures & Algorithms 17 (2000), [4] S. Even and A. Litman, Layered cross product: a technique to construct interconnection networks. Networks 29 (1997), [5] T. Feder and M. Vardi, Monotone monadic SNP and constraint satisfaction, 25th STOC [6] T. Feder and M. Vardi, The computational structure of monotone monadic SNP and constraint satisfaction: a study through Datalog and group theory, SIAM J. Comput. (28) (1998) [7] P. Hell and J. Nešetřil, On the complexity of H-coloring, Journal of Combinatorial Theory Series B 48 (1990), [8] P. Hell and J. Nešetřil, Graphs and homomorphisms, Oxford University Press, [9] W. Imrich and S. Klavžar, Product graphs: structure and recognition, Wiley, New York, [10] L. Lovász, Operations with structures, Acta. Math. Acad. Sci. Hung., 18 (1967), [11] J. van Lint and R. Wilson, A course in combinatorics (2nd ed.), CUP, Cambridge, ECCC ISSN ftp://ftp.eccc.uni-trier.de/pub/eccc ftpmail@ftp.eccc.uni-trier.de, subject help eccc